Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A system for medical image segmentation, comprising: an input module, configured to provide a plurality of patches of an N-dimensional medical image to be segmented; a computing device, configured to implement a trained first neural network as a first controller network and a trained second neural network as a second controller network, wherein the first controller network is configured to sequentially receive input data representing each of the plurality of patches in a first patch sequence, and to sequentially generate and output data indicative of a respective first patch segmentation mask candidate for each of the plurality of patches, wherein the second controller network is configured to sequentially receive input data representing each of the plurality of patches which are the same, in a second patch sequence, and to sequentially generate and output data indicative of a respective second patch segmentation mask candidate for each of the plurality of patches, and wherein the second patch sequence is different from the first patch sequence; and a memory, shared by the first controller network and the second controller network, wherein the first controller network is further configured to write data relating to a state of the first controller network to the memory, wherein the second controller network is configured to read at least part of the data written by the first controller network from the memory and to utilize the data read, upon generating data indicative of at least one of the second patch segmentation mask candidates, and wherein the computing device is further configured to generate, based on the data indicative of the first and second patch segmentation mask candidates, a final image segmentation mask for segmenting the N-dimensional medical image.
Medical image segmentation involves identifying and delineating structures within medical images, which is crucial for diagnosis and treatment planning. Traditional segmentation methods often struggle with accuracy and efficiency, particularly when processing high-dimensional medical images. This invention addresses these challenges by using a dual-neural-network system that improves segmentation performance through shared memory and sequential processing. The system includes an input module that provides patches of an N-dimensional medical image to be segmented. A computing device implements two trained neural networks: a first controller network and a second controller network. The first controller network processes the patches in a specific sequence, generating segmentation mask candidates for each patch. The second controller network processes the same patches but in a different sequence, also generating segmentation mask candidates. Both networks share a memory, allowing the second network to read and utilize data written by the first network, such as its internal state, to enhance its segmentation accuracy. By combining the outputs of both networks, the system generates a final segmentation mask for the medical image. This approach leverages the complementary strengths of the two networks, improving segmentation quality and robustness. The shared memory mechanism ensures that the networks can learn from each other's processing, leading to more accurate and reliable segmentation results.
2. The system of claim 1 , wherein the second controller network is further configured to write data relating to a state of the second controller network to the memory, and wherein the first controller network is configured to read at least part of the data written by the second controller network from the memory and to utilize the data read, upon generating data indicative of at least one of the first patch segmentation mask candidates.
This invention relates to a system for generating patch segmentation mask candidates in a dual-controller network architecture. The system addresses the challenge of efficiently sharing state information between two controller networks to improve segmentation mask generation. The first controller network generates segmentation mask candidates, while the second controller network manages and processes additional data. The second controller network writes state data to a shared memory, which the first controller network reads to inform its segmentation mask generation process. This shared memory allows the first controller network to access relevant state information from the second controller network, enabling more accurate and context-aware segmentation mask candidates. The system ensures seamless communication between the two networks, enhancing the overall performance of segmentation tasks. The shared memory acts as an intermediary, storing and retrieving data to facilitate coordination between the networks. This approach improves efficiency and accuracy in generating segmentation masks by leveraging shared state information.
3. The system of claim 2 , wherein the second patch sequence equals the first patch sequence, reversed.
A system for processing digital signals includes a first patch sequence generator that creates a first patch sequence from a digital signal, and a second patch sequence generator that creates a second patch sequence from the same digital signal. The second patch sequence is an exact reverse of the first patch sequence. The system also includes a comparator that compares the first and second patch sequences to detect anomalies or errors in the digital signal. The patch sequences are derived from overlapping segments of the digital signal, where each segment is divided into smaller sub-segments, and the patch sequences are generated by combining values from these sub-segments. The system may be used in applications requiring error detection, such as data transmission, storage, or signal processing, where identifying discrepancies between the original and reversed sequences can reveal corruption or tampering. The comparison process may involve analyzing the sequences for mismatches, inconsistencies, or deviations from expected patterns, allowing for real-time or post-processing validation of signal integrity.
4. The system of claim 2 , wherein the computing device is configured to implement at least one of at least one trained third neural network as a first memory write head usable by the first controller network for writing to the memory and at least one trained fourth neural network as a second memory write head usable by the second controller network for writing to the memory.
This invention relates to neural network-based systems for memory management, specifically addressing the challenge of efficiently controlling memory access and writing operations in neural networks. The system includes a computing device with a memory and at least two controller networks that manage memory operations. The computing device is configured to implement at least one trained third neural network as a first memory write head, which the first controller network uses to write data to the memory. Additionally, the computing device can implement at least one trained fourth neural network as a second memory write head, usable by the second controller network for writing to the memory. These neural network-based write heads enable dynamic and adaptive memory management, improving efficiency and performance in neural network operations. The system may also include a trained first neural network as a first memory read head for the first controller network and a trained second neural network as a second memory read head for the second controller network, allowing for coordinated read and write operations. The use of neural networks for memory access control enhances flexibility and adaptability in handling complex data storage and retrieval tasks.
5. The system of claim 1 , wherein the second patch sequence equals the first patch sequence, reversed.
This invention relates to a system for processing sequences of data patches, particularly in applications like data compression, encryption, or signal processing where sequence manipulation is critical. The system addresses the challenge of efficiently handling and transforming sequences of patches to achieve desired outcomes, such as error correction, pattern recognition, or data reconstruction. The system includes a first patch sequence and a second patch sequence, where the second sequence is a reversed version of the first. This reversal operation ensures that the system can process data in both forward and backward directions, which is useful for tasks like bidirectional error correction, symmetric encryption, or time-reversed signal analysis. The patches may represent discrete data units, such as blocks of compressed data, encrypted segments, or signal samples, depending on the application. By reversing the second sequence, the system enables symmetric processing, where operations performed on the first sequence can be mirrored or inverted on the second sequence. This is particularly valuable in applications requiring consistency between forward and backward passes, such as in lossless data compression algorithms or cryptographic protocols that rely on reversible transformations. The system may also include additional components, such as a processor to execute the reversal operation or a memory to store the sequences, ensuring efficient and accurate manipulation of the data patches.
6. The system of claim 1 , wherein the computing device is configured to implement at least one of at least one trained third neural network as a first memory write head usable by the first controller network for writing to the memory and at least one trained fourth neural network as a second memory write head usable by the second controller network for writing to the memory.
This invention relates to neural network-based systems for memory access and control, specifically addressing the challenge of efficiently managing memory writes in complex neural architectures. The system includes a computing device with a memory and at least two controller networks, each responsible for managing memory operations. The computing device implements at least one trained third neural network as a first memory write head, which the first controller network uses to write data to the memory. Similarly, at least one trained fourth neural network functions as a second memory write head, usable by the second controller network for writing to the same memory. These neural networks are specialized for memory access, allowing the controller networks to delegate write operations to them, thereby improving efficiency and reducing the computational burden on the controllers. The system may also include additional components, such as a first read head and a second read head, which are similarly implemented as trained neural networks to handle memory read operations. The use of dedicated neural networks for memory access tasks enables more flexible and optimized memory management, particularly in systems where multiple controllers need concurrent or coordinated access to shared memory resources. This approach enhances performance by leveraging specialized neural networks for specific memory operations, reducing overhead and improving overall system efficiency.
7. The system of claim 6 , wherein the computing device is configured to implement at least one of at least one trained fifth neural network as at least one first memory read head, usable by the first controller network for reading from the memory and at least one trained sixth neural network as at least one second memory read head, usable by the second controller network for reading from the memory, and wherein at least one of the at least one first memory read head and the at least one second memory read head is implemented as a differentiable transformation of context vectors representing the context data within the memory.
The invention relates to a neural network system with multiple controller networks and memory read heads. The system addresses the challenge of efficiently accessing and utilizing context data stored in memory within neural network architectures. The system includes a computing device with at least two controller networks, each responsible for different tasks or operations. The computing device implements at least one trained fifth neural network as a first memory read head, which the first controller network uses to read from the memory. Additionally, the computing device implements at least one trained sixth neural network as a second memory read head, which the second controller network uses to read from the memory. The memory read heads are designed as differentiable transformations of context vectors, allowing the system to process and retrieve context data in a differentiable manner. This approach enhances the system's ability to dynamically access and utilize stored information, improving performance in tasks requiring context-aware decision-making. The differentiable nature of the read heads ensures smooth integration with gradient-based optimization methods, facilitating efficient training and adaptation of the neural network system.
8. The system of claim 7 , wherein the at least one first memory read head and the at least one second memory read head include at least one of a plurality of first memory read heads, and a plurality of second memory read heads.
This invention relates to a memory system with multiple read heads for improved data access. The system addresses the problem of limited data throughput in conventional memory architectures by using parallel read operations. The system includes at least one first memory read head and at least one second memory read head, where these read heads can be configured as a plurality of first memory read heads and a plurality of second memory read heads. The read heads operate in parallel to simultaneously access different memory locations, increasing data retrieval efficiency. The system may also include a memory array with multiple memory cells, where the read heads are positioned to independently read data from these cells. The parallel read heads reduce latency and improve bandwidth by allowing concurrent access to multiple data points. This configuration is particularly useful in high-performance computing applications where rapid data access is critical. The system may further include control circuitry to manage the read operations, ensuring synchronized and error-free data retrieval. The use of multiple read heads enhances scalability, allowing the system to adapt to varying data access demands.
9. The system of claim 1 , wherein the computing device is configured to implement at least one of at least one trained fifth neural network as at least one first memory read head, usable by the first controller network for reading from the memory and at least one trained sixth neural network as at least one second memory read head, usable by the second controller network for reading from the memory, and wherein at least one of the at least one first memory read head and the at least one second memory read head is implemented as a differentiable transformation of context vectors representing context data within the memory.
This invention relates to neural network-based systems for memory access, specifically improving how controller networks interact with memory structures. The problem addressed is the inefficiency and inflexibility of traditional memory read mechanisms in neural architectures, which can limit performance and adaptability. The system includes a computing device with a memory and at least two controller networks. Each controller network is assigned a dedicated neural network-based memory read head, implemented as a trained fifth or sixth neural network. These read heads enable the controller networks to access and retrieve data from the memory. The read heads are designed as differentiable transformations of context vectors, which represent context data stored in the memory. This approach allows the read heads to dynamically adapt their memory access patterns based on the context, improving efficiency and accuracy. The differentiable nature of the transformations ensures that the read heads can be integrated into end-to-end learning frameworks, enabling the entire system to be trained in a unified manner. This enhances the system's ability to generalize across different tasks and environments. The use of multiple read heads allows for parallel or coordinated memory access, further optimizing performance. The invention is particularly useful in applications requiring complex memory interactions, such as reinforcement learning, natural language processing, or other AI-driven tasks.
10. The system of claim 9 , wherein the at least one first memory read head and the at least one second memory read head include at least one of a plurality of first memory read heads, and a plurality of second memory read heads.
This invention relates to a memory system with multiple read heads for improved data access. The system addresses the problem of slow data retrieval in conventional memory architectures by using parallel read operations. The system includes at least one first memory read head and at least one second memory read head, where these read heads can be configured as multiple first read heads and multiple second read heads. The read heads operate independently to access different memory locations simultaneously, increasing throughput. The system may also include a memory controller that coordinates the read operations, ensuring efficient data retrieval. The memory may be organized in a way that allows the read heads to access data in parallel without interference. This design is particularly useful in high-performance computing environments where fast data access is critical. The use of multiple read heads reduces latency and improves overall system efficiency by enabling concurrent data retrieval from different memory regions. The system may also include error detection and correction mechanisms to ensure data integrity during parallel read operations.
11. The system of claim 1 , further comprising: a database, shared by the first controller network and the second controller network, the database comprising key vectors linked to a plurality of datasets, wherein the computing device is configured to implement at least one of at least one trained seventh neural network as at least one first database read head usable by the first controller network and at least one trained eighth neural network as at least one second database read head usable by the second controller network, wherein at least one of the at least one first database read head, and the at least one second database read head is configured and usable to retrieve, based on the key vectors, data based on at least one of the plurality of datasets from the database; and wherein a respective one of at least one of the first controller network and second controller network is configured to utilize the data retrieved from the database upon generating data indicative of at least one of the first patch segmentation mask candidate and the second patch segmentation mask candidates, respectively.
This invention relates to a system for processing data using multiple controller networks that share a common database. The system addresses the challenge of efficiently retrieving and utilizing relevant data from a shared database to improve the performance of machine learning models, particularly in tasks like image segmentation. The system includes a database containing key vectors linked to multiple datasets. The database is accessible by both a first and a second controller network. The system employs at least one trained neural network as a first database read head for the first controller network and at least one trained neural network as a second database read head for the second controller network. These read heads are configured to retrieve data from the database based on the key vectors. The retrieved data is then used by the respective controller networks to generate segmentation mask candidates, such as first and second patch segmentation mask candidates. The use of neural networks as database read heads allows for efficient and accurate data retrieval, enhancing the performance of the controller networks in tasks like image segmentation. The shared database ensures consistency and reduces redundancy, while the specialized read heads enable tailored data access for each controller network. This approach improves the overall efficiency and accuracy of the system in processing and analyzing data.
12. The system of claim 1 , wherein the computing device is configured to implement a trained ninth neural network as an encoder module, and wherein the encoder module is configured to receive patches from the input module and to generate the input data representing each of the plurality of patches for the first controller network and the second controller network.
This invention relates to a neural network-based system for processing image data, specifically addressing the challenge of efficiently encoding image patches for downstream machine learning tasks. The system includes a computing device that implements a trained ninth neural network as an encoder module. This encoder module receives image patches from an input module and generates input data representations for each patch. These representations are then provided to two distinct controller networks—a first controller network and a second controller network. The encoder module standardizes the input data, ensuring compatibility with the processing requirements of both controller networks. The first and second controller networks may perform different tasks, such as classification, segmentation, or other image analysis functions, leveraging the encoded patch representations. The system optimizes computational efficiency by reusing the encoder module's output across multiple networks, reducing redundancy and improving performance. This approach is particularly useful in applications requiring multi-task learning or parallel processing of image data, such as autonomous systems, medical imaging, or industrial inspection. The encoder module's design ensures that the input data is transformed into a format that preserves relevant features while minimizing computational overhead.
13. The system of claim 1 , wherein the computing device is further configured to generate, using the data indicative of the first patch segmentation mask candidates, data indicative of a first image segmentation mask candidate, to generate, using the data indicative of the second patch segmentation mask candidates, data indicative of a second image segmentation mask candidate, and to generate data indicative of a final image segmentation mask based on the first image segmentation mask and the second image segmentation mask candidate; and wherein the system further comprises a decoder module configured to generate, from the data indicative of the final image segmentation mask, the final image segmentation mask for segmenting the N-dimensional medical image.
This invention relates to medical image segmentation, specifically improving the accuracy of segmenting N-dimensional medical images such as MRI or CT scans. The system addresses challenges in accurately identifying and delineating anatomical structures or regions of interest within medical images, which is critical for diagnosis, treatment planning, and monitoring. Traditional segmentation methods often struggle with complex structures, noise, or low-resolution data, leading to incomplete or inaccurate results. The system processes an N-dimensional medical image by first dividing it into smaller patches. It then generates multiple segmentation mask candidates for each patch, representing potential segmentations of the image regions. These patch-level candidates are combined to produce initial image-level segmentation mask candidates. The system further refines these candidates to generate a final image segmentation mask, which accurately delineates the target structures within the medical image. A decoder module then converts the final mask data into a usable segmentation output, providing precise boundaries for medical analysis. The approach leverages patch-based processing to enhance segmentation accuracy, particularly in complex or low-contrast regions, while ensuring the final mask maintains coherence across the entire image. This method improves reliability in medical imaging applications where precise segmentation is essential.
14. The system of claim 13 , wherein the computing device is further configured to implement a trained tenth neural network as the decoder module.
The system relates to neural network-based data processing, specifically improving decoding efficiency in machine learning models. The problem addressed is the computational overhead and inefficiency in traditional decoder modules, which can slow down inference and training processes. The invention provides a system where a computing device implements a trained neural network as the decoder module to enhance performance. This neural network is optimized for decoding tasks, reducing latency and resource consumption compared to conventional methods. The system integrates this decoder module into a broader machine learning pipeline, allowing for faster and more accurate data processing. The trained neural network is specifically designed to handle decoding operations, ensuring compatibility with various input data types and improving overall system efficiency. By leveraging advanced neural network architectures, the system achieves superior decoding performance while maintaining accuracy. This approach is particularly useful in applications requiring real-time processing, such as natural language understanding, image recognition, and speech synthesis. The use of a dedicated neural network for decoding tasks ensures that the system remains scalable and adaptable to different machine learning workflows.
15. The system of claim 14 , wherein the computing device is further configured to implement a trained eleventh neural network as a fusion module, and wherein the fusion module is configured to generate the data indicative of the final image segmentation mask based on the data indicative of the first image segmentation mask candidate and the second image segmentation mask candidate.
This invention relates to image segmentation systems that use multiple neural networks to improve segmentation accuracy. The problem addressed is the challenge of generating precise segmentation masks from input images, particularly when relying on a single neural network may lead to errors or inaccuracies. The solution involves a system that employs at least two distinct neural networks to produce separate segmentation mask candidates, which are then combined using a fusion module to generate a final, more accurate segmentation mask. The system includes a computing device configured to process an input image using a first neural network to generate a first segmentation mask candidate and a second neural network to generate a second segmentation mask candidate. These candidates are then processed by a fusion module, which is implemented as a trained eleventh neural network. The fusion module integrates the outputs of the first and second neural networks to produce a final segmentation mask that is more reliable than either individual candidate. The fusion module may use techniques such as weighted averaging, consensus-based merging, or learned feature combination to enhance the final output. This approach leverages the strengths of multiple neural networks while mitigating their individual weaknesses, resulting in improved segmentation performance for tasks such as medical imaging, autonomous driving, or object detection.
16. The system of claim 13 , wherein the computing device is further configured to implement a trained eleventh neural network as a fusion module, and wherein the fusion module is configured to generate the data indicative of the final image segmentation mask based on the data indicative of the first image segmentation mask candidate and the second image segmentation mask candidate.
This invention relates to image segmentation systems that use multiple neural networks to improve segmentation accuracy. The problem addressed is the challenge of generating precise segmentation masks from input images, particularly when relying on a single neural network may produce suboptimal results due to limitations in model architecture or training data. The system includes a computing device configured to process an input image using at least two distinct neural networks to generate separate segmentation mask candidates. A first neural network produces a first segmentation mask candidate, while a second neural network generates a second segmentation mask candidate. These candidates are then combined using a fusion module, which is implemented as an additional trained neural network. The fusion module processes the outputs of the first and second neural networks to produce a final, refined segmentation mask. This approach leverages the strengths of multiple models to enhance segmentation accuracy and robustness. The fusion module may use techniques such as weighted averaging, consensus-based merging, or learned feature combination to integrate the candidate masks into a single, high-quality output. The system is particularly useful in applications requiring precise segmentation, such as medical imaging, autonomous driving, or industrial inspection.
17. The system of claim 1 , wherein all of the neural networks of the system are configured to be differentiable.
A system for neural network-based processing includes multiple neural networks that are all configured to be differentiable. Differentiability allows the system to use gradient-based optimization techniques, enabling efficient training and adaptation of the neural networks. The system may be used in applications such as machine learning, artificial intelligence, or data processing, where differentiable models are required for tasks like backpropagation, parameter optimization, or gradient-based learning. By ensuring all neural networks in the system are differentiable, the system supports seamless integration and end-to-end training of complex models. This differentiability may be achieved through the use of differentiable activation functions, loss functions, and architectural choices that preserve the gradient flow throughout the network. The system may also include mechanisms for handling non-differentiable operations, such as approximation or relaxation techniques, to maintain differentiability across the entire pipeline. This approach enhances the system's ability to learn from data and adapt to new tasks efficiently.
18. A method for medical image segmentation, comprising: providing a plurality of patches of an N-dimensional medical image to be segmented; sequentially receiving, by a first trained neural network acting as a first controller network, in a first patch sequence, input data representing each of the plurality of patches; sequentially receiving, by a second trained neural network acting as a second controller network, input data representing each of the plurality of patches in a second patch sequence, sequentially generating and outputting, by the first controller network, data indicative of a respective first patch segmentation mask candidate for each of the plurality of patches; sequentially generating and outputting, by the second controller network, data indicative of a respective second patch segmentation mask candidate for each of the plurality of patches, the second patch sequence being different from the first patch sequence; writing, by the first controller network, data relating to a state of the first controller network, to a memory for at least the second controller network; reading from the memory, by the second controller network, at least part of the data written by the first controller network; utilizing, by the second controller network, the data read, upon generating data indicative of at least one of the second patch segmentation mask candidates; and generating, based on the data indicative of the first patch segmentation mask candidates and the second patch segmentation mask candidates, a final image segmentation mask for segmenting the N-dimensional medical image.
Medical image segmentation involves partitioning medical images into distinct regions for analysis. Traditional methods often struggle with accuracy and consistency due to variations in image quality, noise, and anatomical differences. This invention addresses these challenges by using multiple neural networks to improve segmentation reliability. The method processes an N-dimensional medical image by dividing it into multiple patches. Two trained neural networks, acting as controller networks, independently analyze these patches in different sequences. The first network generates segmentation mask candidates for each patch in its assigned sequence and writes its internal state data to a shared memory. The second network reads this state data while generating its own segmentation mask candidates in a different patch sequence. By leveraging the shared state information, the second network refines its outputs based on the first network's processing. The final segmentation mask is derived by combining the outputs from both networks, enhancing accuracy and robustness. This approach ensures that the segmentation accounts for contextual information across different patch sequences, improving overall performance.
19. The method of claim 18 , further comprising: writing to a memory, by the second controller network, data relating to a state of the second controller network; reading from the memory, by the first controller network, at least part of the data written by the second controller network; and utilizing, by the first controller network, the data read, upon generating data indicative of at least one of the first patch segmentation mask candidates.
This invention relates to a system for generating patch segmentation masks in a redundant controller network, addressing the challenge of maintaining data consistency and operational continuity between primary and secondary controllers. The system includes a first controller network and a second controller network, where the second controller network operates as a redundant backup to the first. The second controller network generates multiple patch segmentation mask candidates for image processing tasks, such as object detection or segmentation. To ensure reliability, the second controller network writes data relating to its operational state, including the generated mask candidates, to a shared memory. The first controller network then reads this data from the memory and uses it to refine or validate its own patch segmentation mask candidates. This exchange of state data between the two controller networks enhances fault tolerance and ensures that the primary controller can leverage the secondary controller's computations, improving overall system robustness. The method ensures that even if the primary controller experiences disruptions, the secondary controller's data can be utilized to maintain accurate segmentation mask generation.
20. A non-transitory computer-readably data storage medium storing executable program code, configured to, upon the program code being executed on a computer, perform the method of claim 18 .
A system and method for optimizing data processing in a distributed computing environment addresses inefficiencies in task allocation and resource utilization. The invention involves a distributed computing system where tasks are dynamically assigned to processing nodes based on real-time performance metrics, such as processing speed, memory availability, and network latency. The system monitors these metrics across multiple nodes and adjusts task distribution to balance workloads, reducing bottlenecks and improving overall system efficiency. A central controller collects performance data from each node, analyzes it to identify underutilized or overloaded nodes, and reallocates tasks accordingly. The system also includes a predictive model that forecasts future resource demands based on historical data, allowing proactive adjustments before performance degradation occurs. Additionally, the system prioritizes tasks based on urgency and resource requirements, ensuring critical operations are processed first. This approach enhances scalability, reduces processing delays, and optimizes resource usage in large-scale distributed computing environments. The invention is implemented via executable program code stored on a non-transitory computer-readable medium, which, when executed, performs the described method.
21. A non-transitory computer-readably data storage medium storing executable program code, configured to, upon the program code being executed on a computer, perform the method of claim 19 .
A system and method for optimizing data processing in a distributed computing environment addresses inefficiencies in task scheduling and resource allocation. The invention provides a solution for dynamically adjusting computational workloads across multiple nodes to improve performance and reduce latency. The system monitors resource utilization, identifies bottlenecks, and redistributes tasks based on real-time data to balance the load. It employs predictive algorithms to anticipate future resource demands and preemptively allocates resources to prevent delays. The method includes collecting performance metrics from distributed nodes, analyzing the data to detect imbalances, and automatically reassigning tasks to underutilized nodes. Additionally, the system prioritizes critical tasks to ensure timely completion while minimizing idle time across the network. The invention also includes a feedback mechanism that continuously refines the scheduling algorithm based on historical performance data. This approach enhances scalability and efficiency in large-scale distributed systems, particularly in cloud computing and high-performance computing environments. The solution is implemented via executable program code stored on a non-transitory computer-readable medium, ensuring portability and ease of deployment across different hardware configurations.
Unknown
November 3, 2020
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.